Detect faces in a video frame, base on the dimension of the detected face you might want to disregard them and only capture the good ones. Then you need to resize them to a fix size and save them into a mat file of what ever dimension. e.g. h_w_3*num_faces
From #16